Scale space tracking and deformable sheet models for computational vision

The deformable sheet, a physical model that provides a natural framework for addressing many vision problems that can be solved by smoothness-constrained optimization, is described. Deformable sheets are characterized by a global energy functional, and the smoothness constraint is represented by a linear internal energy term. Analogous to physical sheets, the model sheets are deformed by problem-specific external forces and, in turn, impose smoothness on the applied forces. The model unifies the properties of scale and smoothness into a single parameter that makes it possible to perform scale space tracking by properly controlling the smoothness constraint. Specifically, the desired scale space trajectory is found by solving a differential equation in scale. The simple analytic dependence on scale also provides a mechanism for adaptive step size control. Results from application of the deformable sheet model to various problems in computational vision are presented.<
>